@retaw106/mmc-oc 0.0.0-stage → 1.2.2

package/src/shared/jsonc-edit.ts

@@ -0,0 +1,387 @@
+import {
+    applyEdits,
+    createScanner,
+    findNodeAtLocation,
+    getNodeValue,
+    type JSONPath,
+    modify,
+    type Node,
+    type ParseError,
+    parseTree,
+} from "jsonc-parser/lib/esm/main.js";
+
+// ^ Deep ESM import on purpose. jsonc-parser has no "exports" map, and its
+// "main" points at a UMD build whose runtime-relative requires (./impl/format
+// etc.) survive bundling verbatim and then fail to resolve from inside a
+// bundled chunk. When this module became reachable from the plugin entry (the
+// config raw-loader), that broken require made the ENTIRE plugin bundle fail
+// to load at boot. The ESM build bundles statically and safely.
+
+interface TokenLocation {
+    offset: number;
+    length: number;
+    line: number;
+}
+
+/**
+ * jsonc-parser exposes SyntaxKind as a const enum, which cannot be referenced
+ * from this isolated TypeScript module. Keep the scanner token values local.
+ */
+const TOKEN_COMMA = 5;
+const TOKEN_LINE_COMMENT = 12;
+const TOKEN_BLOCK_COMMENT = 13;
+const TOKEN_EOF = 17;
+
+function parseDocument(text: string): Node {
+    const errors: ParseError[] = [];
+    const root = parseTree(text, errors, { allowTrailingComma: true });
+    if (!root || errors.length > 0) {
+        throw new Error("Cannot edit invalid JSONC");
+    }
+    return root;
+}
+
+function findNode(text: string, path: JSONPath): Node | undefined {
+    return findNodeAtLocation(parseDocument(text), path);
+}
+
+function findComma(text: string, start: number, end: number): TokenLocation | undefined {
+    const scanner = createScanner(text, false);
+    scanner.setPosition(start);
+
+    for (;;) {
+        const kind = scanner.scan();
+        const offset = scanner.getTokenOffset();
+        if (kind === TOKEN_EOF || offset >= end) return undefined;
+        if (kind === TOKEN_COMMA) {
+            return {
+                offset,
+                length: scanner.getTokenLength(),
+                line: scanner.getTokenStartLine(),
+            };
+        }
+    }
+}
+
+function lineStart(text: string, offset: number): number {
+    const previousNewline = text.lastIndexOf("\n", offset - 1);
+    return previousNewline === -1 ? 0 : previousNewline + 1;
+}
+
+function lineEnd(text: string, offset: number): number {
+    const nextNewline = text.indexOf("\n", offset);
+    if (nextNewline === -1) return text.length;
+    return text[nextNewline - 1] === "\r" ? nextNewline - 1 : nextNewline;
+}
+
+function inlineCommentEnd(
+    text: string,
+    start: number,
+    line: number,
+    limit: number,
+): number | undefined {
+    const scanner = createScanner(text, false);
+    scanner.setPosition(start);
+    let end: number | undefined;
+
+    for (;;) {
+        const kind = scanner.scan();
+        const offset = scanner.getTokenOffset();
+        if (kind === TOKEN_EOF || offset >= limit || scanner.getTokenStartLine() !== line) {
+            return end;
+        }
+        if (kind === TOKEN_LINE_COMMENT || kind === TOKEN_BLOCK_COMMENT) {
+            end = offset + scanner.getTokenLength();
+        }
+    }
+}
+
+/**
+ * Keep comments written inline after a preceding separator with that preceding
+ * entry. Comments on a later line belong to the entry that follows the comma.
+ */
+function startAfterPrecedingInlineComments(
+    text: string,
+    comma: TokenLocation,
+    entryOffset: number,
+): number {
+    const commentEnd = inlineCommentEnd(text, comma.offset + comma.length, comma.line, entryOffset);
+    if (commentEnd === undefined) return comma.offset + comma.length;
+
+    const nextNewline = text.indexOf("\n", commentEnd);
+    return nextNewline === -1 ? commentEnd : nextNewline + 1;
+}
+
+/** Remove inline comments following a removed entry's comma with that entry. */
+function endAfterFollowingInlineComments(
+    text: string,
+    comma: TokenLocation,
+    nextEntryOffset: number,
+): number {
+    const commentEnd = inlineCommentEnd(
+        text,
+        comma.offset + comma.length,
+        comma.line,
+        nextEntryOffset,
+    );
+    return commentEnd === undefined ? comma.offset + comma.length : lineEnd(text, commentEnd);
+}
+
+function closingWhitespace(text: string, entryEnd: number, closingBracket: number): string {
+    const trailing = text.slice(entryEnd, closingBracket);
+    const finalNewline = trailing.lastIndexOf("\n");
+    if (finalNewline === -1) return "";
+    const newlineStart = trailing[finalNewline - 1] === "\r" ? finalNewline - 1 : finalNewline;
+    return trailing.slice(newlineStart);
+}
+
+function removeArrayEntry(text: string, array: Node, index: number): string {
+    const entries = array.children ?? [];
+    const entry = entries[index];
+    if (!entry) return text;
+
+    const closingBracket = array.offset + array.length - 1;
+    if (entries.length === 1) {
+        const start = array.offset + 1;
+        const replacement = closingWhitespace(text, entry.offset + entry.length, closingBracket);
+        return text.slice(0, start) + replacement + text.slice(closingBracket);
+    }
+
+    if (index === entries.length - 1) {
+        const previous = entries[index - 1];
+        if (!previous) return text;
+        const comma = findComma(text, previous.offset + previous.length, entry.offset);
+        if (!comma) return text;
+        const replacement = closingWhitespace(text, entry.offset + entry.length, closingBracket);
+        return text.slice(0, comma.offset) + replacement + text.slice(closingBracket);
+    }
+
+    const next = entries[index + 1];
+    if (!next) return text;
+    const followingComma = findComma(text, entry.offset + entry.length, next.offset);
+    if (!followingComma) return text;
+
+    const start =
+        index === 0
+            ? array.offset + 1
+            : (() => {
+                  const previous = entries[index - 1];
+                  if (!previous) return entry.offset;
+                  const precedingComma = findComma(
+                      text,
+                      previous.offset + previous.length,
+                      entry.offset,
+                  );
+                  return precedingComma
+                      ? startAfterPrecedingInlineComments(text, precedingComma, entry.offset)
+                      : entry.offset;
+              })();
+    const end = endAfterFollowingInlineComments(text, followingComma, next.offset);
+
+    return text.slice(0, start) + text.slice(end);
+}
+
+function indentationAt(text: string, offset: number): string {
+    const prefix = text.slice(lineStart(text, offset), offset);
+    return /^[\t ]*$/.test(prefix) ? prefix : "";
+}
+
+function inferIndent(text: string, array: Node): string {
+    const firstEntry = array.children?.[0];
+    if (firstEntry) return indentationAt(text, firstEntry.offset);
+
+    const closingBracket = array.offset + array.length - 1;
+    const closingIndent = indentationAt(text, closingBracket);
+    return `${closingIndent}${closingIndent.includes("\t") ? "\t" : "  "}`;
+}
+
+function appendArrayValue(text: string, array: Node, value: unknown): string {
+    const entries = array.children ?? [];
+    const closingBracket = array.offset + array.length - 1;
+    const serialized = JSON.stringify(value);
+
+    if (entries.length === 0) {
+        if (!text.slice(array.offset, closingBracket).includes("\n")) {
+            return text.slice(0, closingBracket) + serialized + text.slice(closingBracket);
+        }
+
+        const indent = inferIndent(text, array);
+        const closingLineStart = lineStart(text, closingBracket);
+        const eol = text.includes("\r\n") ? "\r\n" : "\n";
+        return (
+            text.slice(0, closingLineStart) +
+            indent +
+            serialized +
+            eol +
+            text.slice(closingLineStart)
+        );
+    }
+
+    const lastEntry = entries.at(-1);
+    if (!lastEntry) return text;
+    const trailingComma = findComma(text, lastEntry.offset + lastEntry.length, closingBracket);
+    const isMultiline = text.slice(array.offset, closingBracket).includes("\n");
+
+    if (!isMultiline) {
+        return (
+            text.slice(0, lastEntry.offset + lastEntry.length) +
+            `,${serialized}` +
+            text.slice(lastEntry.offset + lastEntry.length)
+        );
+    }
+
+    const closingLineStart = lineStart(text, closingBracket);
+    const eol = text.includes("\r\n") ? "\r\n" : "\n";
+    const inserted = `${inferIndent(text, array)}${serialized}${trailingComma ? "," : ""}${eol}`;
+    const withValue = text.slice(0, closingLineStart) + inserted + text.slice(closingLineStart);
+
+    if (trailingComma) return withValue;
+    return (
+        withValue.slice(0, lastEntry.offset + lastEntry.length) +
+        "," +
+        withValue.slice(lastEntry.offset + lastEntry.length)
+    );
+}
+
+/**
+ * Replace one JSONC value without reserializing the rest of the document. New
+ * paths use jsonc-parser's structural edit so the original document remains
+ * untouched outside the inserted property.
+ */
+export function setJsoncValue(text: string, path: JSONPath, value: unknown): string {
+    const node = findNode(text, path);
+    if (node) {
+        if (Object.is(getNodeValue(node), value)) return text;
+        const serialized = JSON.stringify(value);
+        return text.slice(0, node.offset) + serialized + text.slice(node.offset + node.length);
+    }
+
+    return applyEdits(
+        text,
+        modify(text, path, value, { formattingOptions: documentFormatting(text) }),
+    );
+}
+
+/**
+ * The line ending and indentation a document already uses, so an inserted
+ * property is laid out like its siblings instead of being written compactly
+ * onto the previous line.
+ */
+function documentFormatting(text: string): {
+    eol: string;
+    insertSpaces: boolean;
+    tabSize: number;
+} {
+    const eol = text.includes("\r\n") ? "\r\n" : "\n";
+    const indent = /\n([ \t]+)\S/.exec(text)?.[1];
+    if (indent?.startsWith("\t")) return { eol, insertSpaces: false, tabSize: 1 };
+    return { eol, insertSpaces: true, tabSize: indent ? indent.length : 2 };
+}
+
+function removeObjectProperty(text: string, object: Node, key: string): string {
+    const properties = object.children ?? [];
+    const index = properties.findIndex((property) => {
+        const propertyKey = property.children?.[0];
+        return propertyKey !== undefined && getNodeValue(propertyKey) === key;
+    });
+    if (index === -1) return text;
+
+    const property = properties[index];
+    if (!property) return text;
+    const closingBrace = object.offset + object.length - 1;
+    const next = properties[index + 1];
+    if (next) {
+        const followingComma = findComma(text, property.offset + property.length, next.offset);
+        if (!followingComma) return text;
+        // Leading and trailing comments remain in their original positions. A
+        // comment previously adjacent to a migrated field therefore survives as
+        // standalone JSONC rather than being discarded with the old key.
+        return (
+            text.slice(0, property.offset) +
+            text.slice(followingComma.offset + followingComma.length)
+        );
+    }
+
+    const previous = properties[index - 1];
+    const trailingComma = findComma(text, property.offset + property.length, closingBrace);
+    if (!previous) {
+        const afterProperty = property.offset + property.length;
+        if (!trailingComma) return text.slice(0, property.offset) + text.slice(afterProperty);
+        return (
+            text.slice(0, property.offset) +
+            text.slice(afterProperty, trailingComma.offset) +
+            text.slice(trailingComma.offset + trailingComma.length)
+        );
+    }
+
+    const precedingComma = findComma(text, previous.offset + previous.length, property.offset);
+    if (!precedingComma) return text;
+    const afterProperty = property.offset + property.length;
+    const withoutProperty =
+        text.slice(0, precedingComma.offset) +
+        text.slice(precedingComma.offset + precedingComma.length, property.offset) +
+        text.slice(afterProperty);
+    if (!trailingComma) return withoutProperty;
+
+    // The trailing comma position is measured in the original string. Removing
+    // only one character before it (the preceding comma) shifts it left by one.
+    const shiftedTrailingComma = trailingComma.offset - 1;
+    return (
+        withoutProperty.slice(0, shiftedTrailingComma) +
+        withoutProperty.slice(shiftedTrailingComma + trailingComma.length)
+    );
+}
+
+/**
+ * Delete an object property without reserializing sibling fields. Comments are
+ * retained as JSONC trivia so key migrations do not erase a user's notes.
+ */
+export function removeJsoncValue(text: string, path: JSONPath): string {
+    const key = path.at(-1);
+    if (typeof key !== "string") return text;
+    const parent = findNode(text, path.slice(0, -1));
+    if (parent?.type !== "object") return text;
+    return removeObjectProperty(text, parent, key);
+}
+
+/**
+ * Remove matching array entries while retaining the exact bytes for survivor
+ * comments and surrounding JSONC regions.
+ */
+export function removeJsoncArrayEntries(
+    text: string,
+    path: JSONPath,
+    shouldRemove: (entry: unknown) => boolean,
+): { text: string; removed: boolean } {
+    let nextText = text;
+    let removed = false;
+
+    for (;;) {
+        const array = findNode(nextText, path);
+        if (array?.type !== "array") return { text: nextText, removed };
+        const index = (array.children ?? []).findIndex((entry) =>
+            shouldRemove(getNodeValue(entry)),
+        );
+        if (index === -1) return { text: nextText, removed };
+
+        const updated = removeArrayEntry(nextText, array, index);
+        if (updated === nextText) return { text: nextText, removed };
+        nextText = updated;
+        removed = true;
+    }
+}
+
+/** Append values to an existing JSONC array without reserializing sibling fields. */
+export function appendJsoncArrayValues(text: string, path: JSONPath, values: unknown[]): string {
+    let nextText = text;
+
+    for (const value of values) {
+        const array = findNode(nextText, path);
+        if (array?.type !== "array") {
+            return setJsoncValue(nextText, path, values);
+        }
+        nextText = appendArrayValue(nextText, array, value);
+    }
+
+    return nextText;
+}

package/src/shared/jsonc-parser.test.ts

@@ -0,0 +1,46 @@
+import { describe, expect, it } from "bun:test";
+
+import { parseJsonc, parseJsoncRecovering } from "./jsonc-parser";
+
+describe("parseJsonc prototype-pollution hardening", () => {
+    it("rejects dangerous keys recursively, including inside arrays", () => {
+        const rejected: string[] = [];
+        const parsed = parseJsonc<Record<string, unknown>>(
+            `{
+                "safe": 1,
+                "constructor": { "hidden": true },
+                "nested": { "prototype": { "hidden": true }, "safe": 2 },
+                "items": [
+                    { "__proto__": { "polluted": true }, "safe": 3 },
+                    { "safe": 4 }
+                ]
+            }`,
+            { onRejectedKey: (path) => rejected.push(path.join(".")) },
+        );
+
+        expect(rejected).toEqual(["constructor", "nested.prototype", "items.0.__proto__"]);
+        expect(Object.hasOwn(parsed, "constructor")).toBe(false);
+
+        const nested = parsed.nested as Record<string, unknown>;
+        expect(nested).toEqual({ safe: 2 });
+        expect(Object.getPrototypeOf(nested)).toBe(Object.prototype);
+
+        const items = parsed.items as Array<Record<string, unknown>>;
+        expect(items).toEqual([{ safe: 3 }, { safe: 4 }]);
+        expect(Object.getPrototypeOf(items[0])).toBe(Object.prototype);
+        expect("polluted" in items[0]).toBe(false);
+    });
+
+    it("preserves hardening when tolerant recovery encounters __proto__", () => {
+        const rejected: string[] = [];
+        const parsed = parseJsoncRecovering<Record<string, unknown>>(
+            '\\{\n"__proto__":{"polluted":true},"cache_ttl":"1h"}',
+            { onRejectedKey: (path) => rejected.push(path.join(".")) },
+        );
+
+        expect(parsed.value).toEqual({ cache_ttl: "1h" });
+        expect(parsed.issues[0]).toMatchObject({ line: 1, column: 1 });
+        expect(rejected).toContain("__proto__");
+        expect(({} as { polluted?: boolean }).polluted).toBeUndefined();
+    });
+});